Introduction

Introduction

Welcome to our new portal, where we will share valuable indicators derived from earnings conference calls transcripts.

About the Database

This initial release features a time series of inflation expectations linked to a recent academic study. Future updates will not only update the indicator but also introduce additional applications to enhance its utility.

Acknowledgements

We want to express our special thanks to Sebastian Manzi and Jeremias Schneider for their excellent collaboration as research assistants on this project.

Inflation expectations

Data from: “Listening to the price-setters: Inferring inflation expectations from synthetic surveys”

Highlights

  • Novel Approach: The paper introduces a method using fine-tuned LLMs to extract inflation expectations from earnings call transcripts, leveraging unstructured text data for richer insights.

  • Methodology: The process involves fine-tuning LLMs, generating inflation-related text completions, and transforming them into quantitative indicators using natural language inference.

  • Complementary Insights: Synthetic surveys provide valuable insights into business inflation perceptions that complement traditional indicators like inflation swaps or professional surveys.

  • Better than NLP Basics: The proposed approach surpasses traditional NLP techniques, such as bag-of-words or sentiment analysis, by capturing the nuanced context and deeper patterns in textual data.

Abstract

We propose a novel method to extract business inflation expectations via synthetic surveys completed by finetuned large language models (LLMs). These models are trained in pseudo-realtime using earnings conference calls’ transcripts. The analysis covers inflation in the US for the sample period 2011-2023. Synthetic surveys anticipate inflation over multiple forecast horizons. The proposed approach is shown to produce information that goes beyond that found in traditional business surveys and market-based indicators. The predictive performance of synthetic surveys cannot be matched applying conventional text processing techniques. Our results demonstrate the value of expanding the set of expectation metrics used in academic and policy analysis of inflation dynamics.

Data

The paper can be download from here.

The data can be download from here.

Monthly report

The following Table shows the latest numbers of our business inflation expectations via synthetic surveys completed by finetuned large language models (LLMs):

Monthly report - January 2025
Inflation expectations in the US
based on Synthetic business surveys (sbs)
sbs index (last - December 2024) Period
89.86 89.9
Notes: Period contains the last 6 months.

In the following Figure, we show a strong association between inflation and synthetic business inflation expectations.